+$5021,& 02'(/ )25 7+( )/825(6&(17 /$03

Camilo Carrillo

Departamento de Ingeniera Elctrica Universidad de Vigo Lagoas Marcosende s/n 36200 Vigo, Spain

Jos Cidrs, Member IEEE

   !#"%$"&&  '"&(0)1" 2 $43 2 %) 2 $"5 5  67  9@AA" 6  6 B "C "$D!EF3'5 9 2 A"&&C" 6 G5 )HI&PH0) 2 $"5#QR' $"CSC)"5 2'# 6"  8 T $ 9  67 "VU!#"  )" 6 5WA"&95 X3 2 )Y 2 )`)1"A  a5 5 E 2 5 "b5 )c S  5 a d $8 e"  6 2 C 2 9 6 c  "f 9'" 6 $g@a5 5 ' 62@6 5  6 "   "&` 6 $8  " 6  G$" #" 6 $ " C" 2'6 2 9@AC"FS 2 5 7 "FS  a  2'6 &P  )195 a  2@6 )" 2 $hQ   6  "ia  5  Ep 2 6 a5 Ea&" 62'6 5  6 "  " Q C e C " "  $ " $ 3 f )  $ " 5 R P "d 2  2 &"$g 6 "VUa " 6  2'6 2 6 "srA2 "a  SC"s6 tF) 2   a5 Ea 2 fu rvtRq1  x w 2 a5 95 a "f  "  'A) 2'6  2'E8 6 35 9 6 A"&&"  b5 )cI&  9@AA" 6   6 B "C "$y! P  "rtsXa5 7H2   )  9@AA" 6 5 E09"$ 2 6 a5 Ea2 &"%  " 6  6 "  2' 6 @" QG"C" 6y62'6 5  6 "  2'6 SC" "!# 6 $b5  6 " 6 " Q 2 Ae&P
, ,1752'8&7,21

every iteration. In order to achieve the greatest rate of convergence, the current across the tube is used as the interface because its waveshape has a distortion lower than the voltage of the tube.
,, '(5,9$7,21 2) 7+( /$03 02'(/

The wide use of the fluorescent lamps for illumination makes necessary the study of their influence in the network harmonic spectrum. The fluorescent tube operation is based on the setting up of an arc into the tube. This arc has a clear non-linear behaviour. In order to get a study based on simulation, first it is necessary to achieve a model that represents the more important lamp non-linearities. We proposed a model that takes into account the tube and ballast non-linearities [1] and their dependence on source voltage variations. A simulation method with the ability to analyse non linear networks is needed for this model. We proposed an extension on the Iterative Harmonic Analysis (IHA) to calculate the harmonic currents injected by fluorescent lamps. The IHA algorithm is widely used to analyse the interaction between non-linear converters and linear networks[2][3][4]. It involves a hybrid technique which formulates the non linear component in the time domain and the linear system in the frequency domain, using the common voltage to interface between the two systems at

Figure 1 is represents the lamp circuit. The model will represent the normal tube operation with the starter opened. Thus, the starter has not been modelled. The source voltage e(t) has a sinusoidal waveshape with the expression: H( W ) = 2 ( ` VLQ( 2)U W ) (1) where Fr is the frequency (50Hz) and ERMS is the RMS source voltage. The voltage and current waveforms of a typical 36W tube operation at different source voltage levels, ERMS={200V,220V,240V}, are illustrated in Figure 2 and Figure 3.

e AfV gAf

Lhist

d hVi j kHlImn o!mpq!r#sat u v&i w xy j k

d
L

d
R

A z{ |}'~A &z }'~AAvA&@ `A'}@{ C A@a{ ~AA}@{

0,6 0,5 200 220 240

0,4 0,3 0,2 0,1 0

0,01 0,005 z{ |}'~A`C 'A~A{  'd&}'~A~AA@aA~Av ' }'@cs{ `@{ A~A&@ V}'~AAC C|@  H HCc R 0

110 100 90 80 70 60 50 40 30 20 10 0 0

200 220 240

0,01 0,005 z{ |}'~&cC 'A~{ ` 'C C|I 'd }''cR{ `@{ A~AA@ v}'~AAC C|d A H Cc  RI

approximate the part of voltage with more level of noise. As can be seen in Figure 6 this area is the wave between W1 and W2. The position of these two points has no dependence on the source voltage level. So the width of )LOWHUHG $UHD is constant even thought ERMS takes different values between 200 and 240V. Moreover, the portions of waveforms between WA and W1 and between W2 and WB are independent on ERMS level. Thess conditions make the implementation of LSQ easy.
120 V Filtered Area

The steady state characteristic of the tube with ERMS = 220V is shown in Figure 4. As can be seen, the curves obtained have an important level of noise. The first step to get an approximate characteristic, which allows a model to be achieved, is a filtering of the voltage and current to focus the main points of the characteristic.
0,6 iT(t) [A]

0 Measured Filtered

sg 0,01

za{ |}'~Ad'#C C|I 'd }''I &~d''A ~Ad 'I { A~{ @|
-110 u(t) [V] +110

A point by point ratio of the magnitudes of the voltage and current waveforms of Figure 5 and Figure 6 yields the steady state non-linear characteristic of the tube shown in Figure 7. As can be seen in this figure, the characteristic is affected by the RMS voltage of the supply voltage (ERMS).

@ 

za{ |}'~Ad}@ {'A'~AA A~{ { d 'd }'~AAVAA@ }''

The current across the tube iT(t) can be easily filtered using a low-pass filter with a cut-off frequency of 1kHz. The result is shown in Figure 5.
0,5 A Measured Filtered

-0,6

ERMS = 220V

0,6

I C I C ad IC

@ '  za{ |}'~Ad'@za{ A~AAA'~A A~{ v { I 'd }''

110

To model this characteristic an approximation by straight traces is used (see Figure 8). Small variations in the supply voltage (ERMS) only affect substantially the position of point E = {UE, IE} that follows the straight line (E1-E2) :
0 0 sg 0,01
8

= 26,9 , + 66,5

(2)

z{ |}'~Ad'' }'~A~AA@aA~AV ' }''d A~I'`'A ~Ad 'I { A~A{ @|
The voltage has more difficulties because it has a harmonic spectrum with high frequencies due to its waveshape. Then, a Least Square Method (LSQ) is the way to eliminate the noise. A 5th grade polynomial is used to

  c
0,6

`G

L W

( ) = ( ) +
D W

2 (W )

(3)

where (t) is the flux and a and b are constants,. The parallel resistance is derived from the hysteresis curve by the expression:
5

110

=2)021

IU

/ G3

(4)

  z{ |}'~Ad's''~A{ C I}@ {'A'~AA A~{ {

where max is the maximum flux, fr the fundamental frequency and di one half of the hysteresis width at =0.
,,, ,7(5$7,9( +$5021,& $1$/<6,6

The behaviour of the u-i characteristic obtained above remains valid for small variations (from 40Hz to 50Hz) in the frequency of power source voltage , Fr, as shown in Figure 9. In order to compare the characteristic behaviour under the two situations, the point E position and its approximation via LSQ (see Figure 8) are represented in Figure 10.

C 
0,7

CCc& CCA C& @

Having obtained accurate models for the different elements of the lamp, it is necessary to use an iterative algorithm to derive the tube voltage and the current harmonic spectra. The IHA algorithm shown in Figure 13 is used for this purpose. In a first approximation a tube equivalent resistance is used for the calculation of initial currents. This resistance is derived from the values of fundamental frequency voltage and current obtained from tests, with small current in the starting capacitor (ic(t)) ignored. The main blocks in the algorithm are the Calculation of Tubes Voltage and the Correction of the Characteristic. In order to do this correction, the maximum of the current across the tube iT(t) is obtained, so:
,

&  0 120 za{ |}'~Ad'@z{ A~AAA'~A A~{ v { I 'd }''ICa@{ A~AA@ ~AA'}'A'{ A
56 55

4 =

PD[ L

{ 5 ( )}
W

(5)

"! #"!
E E-LSQ

a 
E E-LSQ

R

54 53 52 51

and then the value of UE is calculated from Equation 2. With the corrected characteristic, the tube voltage is derived from the approximate characteristic as shown in Figure 14. Another important block is the calculation of the current through the reactor that is derived as shown in Figure 15; it is first necessary to calculate the flux from:

E {UE, IE}

(W ) =

GY

()
W

GW

(6)

R  z{ |}'~A#&'#~A{ C { `'C{ @g'{ { `'%$'&"( ''~A{ `C { @

0,46 0,50 0,54 0,58 0,62 0,66 0,70

50 0,42

At this point the most important non-linearity of the tube is modelled. The other important non-linearity is the ballast due to the presence of some saturation and hysteresis in the iron-core ballast. These effects are modelled with a parallel inductance/resistance equivalent The inductance has the expression[5]:

which is solved via a double FFT transformation. The flux is then used to obtain the i(t) waveform. In the final block a comparison is made with the current obtained in the previous iteration to quantify the error (), ie

= L7
8
=1

( )
Q7

1 Q7

( )

(7)

In Figure 11 and Figure 12 a comparison between measured and simulated waveforms is shown.
0,6 Measured Simulated

120 Measured Simulated

V
A

0
0 sg 0,01

@BA CD"EFHGGI"PRQSUT"VEA WXQYa`bFc d%FFY%c e"F'WXA SaDbf Vc FgUVY"gUSaFVWXD"EFg h D"EEFYbciV h EQWXWc e"F'c D"`"Fqp rtsutvwaxxyR

@BA CD"EFHGxIbPRQSaT"VEA WXQYU`"Fc d%FFYUc e"FqW2A SUDbf Vc FgUVYbgaSUFVWXD"EFg Qf c VCF'Qic e"Fc D"`"F'p r sutv waxxbyq

sg

0,01

Start

Calculation of initial currents

vR(t) = Ri(t)

' d "
Correction of the characteristic. Calculation of tubes voltage

Ic() = jU()

d
-

f hgji f kl f m

Calculation of the current across the ballast

Update i(t) and u(t) No

Error < 1e-5? Si End

@BA CD"EFHGnI"obA SaDbf Vc A QYaQp f QdqgbA VCEVSU

 2  2

rs

rvuxw2y{z z~ |} rqt

@BA CD"EFHG"IFEA Vc A QYaQ Qf c VCFA Y%c e"Fc D"`"F

b ba U ' ji aji b ba U  
vL(t) FFT j IFFT (t)
t i(t) (t) (t)

1/Rm iRm(t)
i(t)

i(t)

iLsat(t)

@pA CD"EFHG"IHFEA Vc A QYaQc e"F h D"EEFYbciA YUc e"FqEFV h c VY h F


,9 7(67 6<67(0 $1' 5(68/76

400 300

When the fluorescent lamp includes a power factor correction capacitor at the input the equivalent circuit will also include a Thevenin equivalent source as shown in Figure 17. The test system illustrated in this figure consists of the lamp components of the previous section, a 220VRMS voltage source in series with a resistive (R=0,37) inductive (L=42H) impedance and a power factor correction capacitor of Cc = 5F. In the example, the equivalent circuit is assumed to be feeding 100 lamps. A second iterative loop[7], shown in Figure 16, is needed to represent the effect of the power factor correction capacitor and the source impedance. The iteration results can be seen in Figure 18 and Figure 19.
Start Calculation of Initial Conditions u0(t), il0(t), ica0(t) = 0 u(t), il(t) Resolution of Non-Linear Circuit

200 100 0 0,00 -100 -200 -300

 
0,01


0,02

@BA CD"EFHG"IQf c VCF'Q h QSaT"FY"WXVc A QY h VT"V h A c QE'VY"gUc D"`"FqQ`bc VA Y"Fg g"D"EA YC{c e"FqW2A SUDbf Vc A QYb
0,60 0,40 0,20 0,00 0,00 -0,20 -0,40

-400


0,01 0,02

"RX  "p X
il(t) Resolution of Linear Circuit e(t)

@BA CD"EFHG"IFc d%QE{VY"gUf VSaT h D"EEFYbc WQ`bc VA Y"FgUg"D"EA YbCc e"F W2A SaDbf Vc A QYb
9 &21&/86,216

-0,60

u(t) No Error < 1e-3? Yes End

@BA CD"EFHG"I"obA SaDbf Vc A QYaQif VSaTUVY"gUY"Fc d%QE

The influence of source voltage level on the characteristic of the tube must be taken into account during the fluorescent lamp simulation. With this consideration, very little error between the measured and simulated waveforms is achieved. The model obtained is valid when small variations in voltage frequency are present. This could be used for a flicker simulation. The presence of noise in the voltage and current waveforms makes necessary a filtering process to analyse the behaviour of the u-i characteristic of the tube. Paying special attention to the filtering process of the voltage in the tube because of its sharpness waveform. So, an LSQ method is needed to filter this voltage.

Lr

Rr
CC

 X 
L RL

i  

CC

CA

CA



@BA CD"EFRG"I"PRQSUTbf Fc FqF"DbA Vf FYbc h A E h DbA cpQic ebF'c FW2cpW "W2c FS%

9, 5()(5(1&(6

[1] J.A. Waymouth, Electric discharge lamps, The MIT Press, 1971. [2] R. Yacamini & C. de Oliveira, Harmonics in multiplie converter systems: a generalised approach, Proc. IEE, V. 127, No 2, March 1980: 96-106. [3] J. Arrillaga, N. R. Watson, J. F. Eggleston, C. D. Callghan, Comparison of steady state and dynamic models for the calculation of ac/dc systems harmonics Proc IEE, V. 134C, No 1, January 1987:31-37. [4] N. R. Watson, A. J. Robbie & J. Arrillaga, Representing transformer saturation in iterative harmonic analysis Proc. Int. Conf. on Harmonics in Power Systems (ICHPS-VI), Bologna, Sept. 1994:318-324. [5] J. Arrillaga, A. Bradley & P.S. Bodger, Power system harmonics, John Wiley & Sons, 1989. [6] Electromagnetic Transient Program (EMTP) Reference Manual. [7] J. Arrillaga, J. Cidrs, C. Carrillo, An iterative algorithm for the analysis of fluorescent lamps ICHPQ, Las Vegas, Oct, 1998: 687-692.
9,, $&.12:/('*(0(17

The financial support received from the Xunta de Galicia under the contract XUGA-32105B95 is gratefully acknowledged by the authors.
9,,, %,2*5$3+<

received his degree in electrical engineering from the University of Las Palmas de G.C. (Spain). He obtained a PhD in electrical engineering from the University of Santiago (Spain) in 1987. He has been a professor at the University of Vigo (Spain) since 1982. He is a Member of IEEE. received his degree in electrical engineering from the University of Vigo (Spain), where he has been a lecturer since 1992.

bQWXqPqA gbEW

PRVS%A f QhPRVEEA f f Q